Generic covers branched over {xn=ym}

Generic covers branched over {xn=ym}

Topology and its Applications 103 (2000) 1–31 Generic covers branched over {x n = y m } Sandro Manfredini ∗ , Roberto Pignatelli 1 Dipartimento di Ma...

295KB Sizes 0 Downloads 101 Views

Topology and its Applications 103 (2000) 1–31

Generic covers branched over {x n = y m } Sandro Manfredini ∗ , Roberto Pignatelli 1 Dipartimento di Matematica, Università di Pisa, Via F. Buonarroti 2, 56127 Pisa, Italy Received 29 September 1997; received in revised form 6 October 1998

Abstract In this paper the authors study generic covers of C2 branched over {x n + y m } = 0 s.t. the total space is a normal analytic surface. They found a complete description of the monodromy of the cover in terms of the monodromy graphs and an almost complete description of the local fundamental groups in case (n, m) = 1. For the general case, they give explicit descriptions of base changes in terms of monodromy graphs; they describe completely the embedded resolution graphs in the case n|m. Via these base changes every cover is a quotient of such a cover.  2000 Elsevier Science B.V. All rights reserved. Keywords: Normal surface singularities; Branched covering spaces; Monodromy AMS classification: 32S25; 32S05

0. Introduction In this paper we study normal singularities of complex analytic surfaces. Recall that (see, e.g., [11]) the singularities of a normal analytic variety form an analytic subvariety of codimension at least 2. So, a normal curve is automatically smooth, and a normal surface is automatically a surface with isolated singularities. The converse holds only for hypersurfaces (see [5]). Then, in case of dimension 2, in order to study germs of normal analytic surfaces we can consider analytic surfaces with just one singular point. Recall that, by Weierstrass preparation theorem, in a suitable neighborhood of every point of an analytic surface there exists a holomorphic function to a disc which is an analytic cover branched over a curve (see [6]). Moreover, a generic function like this one is a “generic” cover, i.e., a branched cover of degree d such that the fiber over a smooth point ∗ Corresponding author. Email: [email protected]. 1 Email: [email protected].

0166-8641/00/$ – see front matter  2000 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 6 - 8 6 4 1 ( 9 8 ) 0 0 1 6 0 - 6

2

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

of the branching curve has d − 1 points (see [12]). Every element in the fundamental group of the set of regular values of this map, induces a permutation of the d points of the fiber over the base point, thus a homomorphism from this group to Sd , called “the monodromy of the cover”. The “generic” condition means that for each geometric loop (i.e., a loop around a smooth point of the curve) its monodromy is a transposition. This property can be usefully applied to study singularities; in fact, given a curve C contained in a disc ∆ (or in C2 ), and a homomorphism µ : π1 (∆ − C) (respectively π1 (C2 − C)) → Sd , s.t. the images of the geometric loops are transpositions, there exists a unique normal surface S and a generic cover from S to ∆ (respectively C2 ) with C as branch locus and µ as monodromy (for an explicit construction see, e.g., [12]; unfortunately, this construction is quite involved, so it does not give directly a satisfying description of normal singularities). So, in order to classify generic covers π : S → C2 of degree d, with S normal, branched over some curve C, we need to classify only the generic monodromies µ : π1(C2 − C) → Sd . In this paper we restrict to the case where the branching curve has (up to analytic equivalence) the equation {x n = y m }. This is a very particular case, but, by the classification of singularities of plane curves given by Puiseux (see [4]), it seems to be the natural starting point. In Section 1 we state some well known expressions of the fundamental group of the disc minus our curves, via generators and relations, and we give a combinatorial bound for the degree of the cover. In Section 2 we prove that our family of covers is stable under base change with maps of type fa,b (x, y) = (x a , y b ); we represent the monodromy of a generic cover of degree d branched on the curve {x n = y m } by a graph with d vertices and n labeled edges and we describe the action of a base change as above over these graphs. In Section 3, we restrict ourselves to the case (n, m) = 1, and we give a complete classification of the graphs associated to these covers. In particular, we prove the following Theorem 0.1. The monodromy graphs for generic covers π : S → C2 of degree d > 3 branched over the curve {x n = y m }, with (n, m) = 1, are the following: (1) “Polygons” with d vertices, valence n/d (or m/d) and increment j , with (j, d) = 1, j < d/2, j (d − j )|m (respectively j (d − j )|n). Moreover, d must divide n (respectively m). (2) “Double stars” of type (j, d − j ) and valence n/j (d − j ) (or m/j (d − j )), with (j, d) = 1, j < d/2, j (d − j )|n (respectively j (d − j )|m). Moreover, d must divide m (respectively n). The base change induced by the map f (x, y) = (y, x) in C2 takes graphs of type 1 in graphs of type 2, and vice versa. For the definition of “polygons” and “double stars” see Definitions 3.2 and 3.3. So, in order to classify generic covers, we get the following

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

3

Corollary 0.2. If (n, m) = 1 then the generic covers π : S → C2 of degree greater than 3 branched over {x n = y m }, S normal surface, are classified by the disjoint union of the sets  (j, d)|d > 3, d|n, (j, d) = 1, j < 12 d, j (d − j )|m and  (j, d)|d > 3, d|m, (j, d) = 1, j < 12 d, j (d − j )|n . In both cases, d is the degree of the cover. All these graphs correspond to some cover also if (n, m) > 1, but they do not give a complete classification. We found also explicit equations for the singularities of S in the case j = 1. In Section 4 we compute the local fundamental group of the surfaces associated to some of the graphs constructed in Section 3. This gives partial answer to the smoothness problem (is S smooth?); moreover it provides an useful tool in the proof of Theorem 0.3. In Section 5 we describe completely the embedded resolution graphs of all the possible singularities in case m = bn for some b, using, to simplify the calculations, the equation {x n + y bn = 0} for the branching locus; we prove the following Theorem 0.3. Let π : S → C2 be a d-sheeted generic cover branched over the curve {x n + y bn = 0}, S normal. S has a resolution which is the plumbing variety of the following normal crossing configuration of smooth curves:

where the vertex E b has genus (n − d − ν + 2)/2 and self-intersection −ν. Moreover, S is smooth ⇔ the monodromy graph is a tree; this can occur only if d divides b. Remark that by the results about base change of Section 2, all the possible surfaces under consideration are the quotient of one of these singularities by the action of a finite group.

1. Fundamental groups and maps Let Cn,m be the curve in C2 defined by the equation x n = y m .

4

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

The aim of this section is to compute some useful presentations for the fundamental group of C2 \ Cn,m and derive the first necessary conditions for the existence of generic covers of given degree branched over Cn,m . Let β, µ¯ j be the paths β(t) = (1, tε +(1−t)(1−ε)), µ¯ j (t) = (1, εe2πitj/m ) for t ∈ [0, 1] and if z ∈ C∗ and λ = (λ1 , λ2 ) is a path in C2 let z(λ) = (λ1 , zλ2 ). Let µ1 , . . . , µm be the geometric basis of π1 ({x = 1} \ Cn,m ) with (1, 1 − ε) (0 < ε < 1/2) as base point given by  µ1 = 1, εei(2πt +π) + 1 t ∈[0,1],  −1 µj = β · µ¯ j · ωj −1 β −1 · µ1 · β · µ¯ −1 j ·β

for j = 2, . . . , m, where ω = e2πi/m .

Proposition 1.1. The fundamental group of C2 \ Cn,m admits the three equivalent presentations (1) π1 (C2 \ Cn,m ) = hµ1 , . . . , µm | µi = Mµi+n M −1 , i = 1, . . . , mi, where M = µ1 · · · µn and the indices are taken to be cyclical (mod m); −1 (2) π1 (C2 \ Cn,m ) = hγ1 , . . . , γn | γi = Γ γi+m Γ , i = 1, . . . , ni, where Γ = γ1 · · · γm and all indices are taken to be cyclical (mod n); (3) π1 (C2 \ Cn,m ) = hγ1 , . . . , γn | Γ = γi · · · γm+i−1 , i = 2, . . . , ni, where all indices are taken to be cyclical (mod n). Proof. Applying Zariski–Van Kampen theorem (see [14,13,9]) to the projection φx : C2 → {y = 1}, (x, y) 7→ (x, 1), we get the first presentation. Let Mj = µ1 · · · µj (with cyclical indices (mod m) and M0 = 1) and define γi = −1 Mi−1 µ−1 i Mi−1 for i = 1, . . . , n. −1 for i = 1, . . . , m, where Γh = γ1 · · · γh (with cyclical indices Since µi = Γi−1 γi−1 Γi−1 (mod n) and Γ0 = 1), γ1 , . . . , γn are a new set of generators and rewriting the relations of the first presentation in terms of the γ ’s we get the second one. The third presentation is easily obtained from the second one. 2 Call µ1 , . . . , µm the standard generators of π1 (C2 \ Cn,m ) and γ1 , . . . , γn the minimal standard generators of π1 (C2 \ Cn,m ). From these presentations it is immediate to verify that, setting Γ = γ1 · · · γn Γ = M −1

and Γ m/(n,m) = Γ

n/(n,m)

is in the center.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

5

Observe that if we apply Zariski–Van Kampen theorem to the projection φy : C2 → {x = 1}, φy (x, y) = (1, y), and proceed as in Proposition 1.1, we may take as generators of π1 (C2 \ Cn,m ) a geometric basis µ˜ 1 , . . . , µ˜ n of π1 ({y = 1} \ Cn,m ) with relations µ˜ i = µ˜ 1 · · · µ˜ m µ˜ i+m (µ˜ 1 · · · µ˜ m )−1 for i = 1, . . . , n and cyclical indices (mod n). Note also that this is the same as calculating the fundamental group of the complement of Cm,n = {x m = y n } via φx . Proposition 1.2. With the above notations µ˜ i = γi . Proof. Let fn,m : C2x,y → C2ξ,η be the map (ξ, η) = fn,m (x, y) = (x n , y m ).   fn,m |C2 \({xy=0}∪Cn,m ) : C2 \ {xy = 0} ∪ Cn,m → C2 \ ξ η(ξ − η) = 0 is a covering. Take (1, 1 − ε) as base point in C2 \ ({xy = 0} ∪ Cn,m ) and take as generators of π1 (C2 \ ({xy = 0} ∪ Cn,m )) the standard generators of π1 (C2 \ Cn,m ) µ1 , . . . , µm , µx = µ¯ m = (1, (1 − ε)e2πit ) (loop around the x-axes) and µy = (e2πit , 1 − ε) (loop around the y-axes) for t ∈ [0, 1]. Observe that if we quotient π1 (C2 \ ({xy = 0} ∪ Cn,m )) by the subgroup normally generated by µx and µy we obtain π1 (C2 \ Cn,m ) and that we can do the same thing with the µ’s ˜ as generators. In the plane ξ + η = 1 + (1 − ε)m = 2 − ε0 take as generators of π1 (C2 \ {ξ η(ξ − η) = 0}) γ , γξ , γη as shown in the figure below where the line ξ + η = 2 − ε0 is identified with C via the η coordinate and p = (1, 1 − ε0 ).

γ , γξ , γη are related by the equations γξ γη γ = γ γξ γη = γη γ γξ . Since (fn,m )∗ is injective, we can identify π1 (C2 \ ({xy = 0} ∪ Cn,m )) with its image obtaining: j −1

µj = γ ξ

−j +1

γ γξ

for j = 1, . . . , m,

µx = γξm , µy = γηn , µ˜ j = γη−j +1 γ −1 γηj −1 µ˜ x = γξ−m , µ˜ y = γη−n .

for j = 1, . . . , n,

6

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Note now that −1 γη−1 γ −1 γη = γ γξ γ −1 γξ−1 γ −1 = µ1 µ−1 2 µ1 ,

thus µ˜ j = γη−j +1 γ −1 γηj −1 = (γ γξ )j −1 γ −1 (γ γξ )−j +1 −1 = µ1 · · · µj −1 µ−1 j (µ1 · · · µj −1 )

as we wanted.

2

Now we look for which d > 3 there exists a surjective homomorphism ρ from the fundamental group of C2 \ {x n = y m } in Sd sending the geometric generators in transpositions, i.e., a normal surface S and a d-sheeted generic covering π : S → C2 branched over {x n = y m }. For every σ ∈ Sd call ν(σ ) the number of orbits of σ , ν the value of this function for ρ(Γ ). ν is the number of cycles of a permutation constructed multiplying n transpositions in Sd , so n + d + ν is even. Call a1 , . . . , aν the length of the ν cycles of the monodromy of Γ . Being Γ m/(m,n) central and d > 3, the order of ρ(Γ ) divides m/(m, n); then every ai does. Let Dmn ⊂ N, be the set of (positive) divisors of m/(m, n), and consider all the possible ways to write d as sum of elements of this set. d be the set of all the possible “lengths” of this sums, where the “length” of a Let Kmn sum, is the number of integers we are adding. Now define the function  d | n + d + ν is even . Λ(m, n, d) = inf ν ∈ Kmn Λ gives a lower bound for the number of orbits of ρ(Γ ); in fact it is the minimal number d ) and with the same of orbits for permutations with order that divides m/(m, n) (ν ∈ Kmn parity of ρ(Γ ) (product of n transpositions). Now define χ(m, n, d) = n + 2 − d − Λ(m, n, d). We have the following Proposition 1.3. Let (S, ϕ) be a generic cover branched on {x n = y m } of degree d > 2. Then χ(m, n, d) > 0. Moreover, if n divides m, the converse holds. Proof. For the first part of the proposition we prove, by induction on n, that for the product of a transitive set of n transpositions in Sd , the number of orbits ν must be ν 6 n + 2 − d. In fact, if n = 1 then d = 2 and ν = 1, and there is nothing to prove. If n > 1, consider the first n − 1 transpositions ρ(γi ), and let ν 0 be the number of orbits of their product; ν 0 = ν + 1 or ν 0 = ν − 1.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

7

If these n − 1 transpositions generate Sd then ν 6 ν 0 + 1 6 n − 1 + 2 − d + 1 = n + 2 − d. Otherwise they generate Sd−k × Sk , and the last transposition “connects” two different orbits of their product. Suppose that in Sk there are exactly g among the first n − 1 transpositions, then their product has by induction ν 00 6 g + 2 − k orbits and the product of the other n − 1 − g transpositions has ν 0 − ν 00 6 n − 1 − g + 2 − (d − k) orbits. So ν = ν 0 − 1 6 g + 2 − k + n − 1 − g + 2 − (d − k) − 1 = n + 2 − d. By definition Λ(m, n, d) 6 ν and we get the result. For the converse, remark that in this case the fundamental group is generated by γ1 , . . . , γn with the only relation that Γ is central. So it is sufficient to exhibit a set of n transpositions σ1 , . . . , σn s.t. their product is the identity of Sd . Now assume χ(m, n, d) > 0, i.e., Λ(m, n, d) 6 n − d + 2. Then there exist a1 , . . . , aΛ(m,n,d) ∈ N such that: P (1) ai = d, (2) ∀ i ai divides m/(m, n), (3) n + d + Λ(m, n, d) is an even number. Choose the following transpositions:    (1, i + 1), 1 6 i 6 d − 1,    !  i−d+1  X σi = 1, d + 1 − ak , d 6 i 6 d + Λ − 2,    1    (1, 2), d + Λ − 1 6 i 6 n. Of course this choice verifies our condition, then it describes a d-sheeted generic cover branched on the curve {x n = y m }. 2 In general these functions are not so simple to compute. The following holds: Remark 1.4. χ(m, n, d + 1) > χ(m, n, d) − 2. Proof. Let Λ = Λ(m, n, d). If Λ = +∞ there is nothing to prove. Otherwise, let a1 , . . . , aΛ ∈ Dmn realizing the minimum as in the definition of Λ. P ai = d + 1 Then a1 , . . . , aΛ , aΛ+1 = 1 ∈ Dmn , with n + d + 1 + Λ + 1 even and Λ+1 1 which implies Λ(m, n, d + 1) 6 Λ(m, n, d) + 1. But then χ(m, n, d + 1) = n + 2 − d − 1 − Λ(m, n, d + 1) > n − d − Λ(m, n, d) = χ(m, n, d) − 2.

2

Remark that if m = n, m even, Λ(m, m, d) = d, i.e., there exist generic covers if and only if m + 2 − 2d > 0, i.e., d 6 m/2 + 1 (if m = n, m odd, there are no generic covers for d > 2; in fact this is true for every n, m s.t. mn is odd, see next section for details).

8

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

2. Fiber products, monodromies and graphs Let π : S → C2 be a d-sheeted generic cover branched over {x n = y m } with S an irreducible surface with an isolated singularity in P = π −1 (0, 0) and let ρ be its monodromy. Consider the map fa,b : C2 → C2 and let S = S ×C2 C2 be the fiber product. We get the following commutative diagram: S

ψ

π¯

C2

S π

fa,b

C2

Proposition 2.1. S as above is an analytic surface with an isolated singularity in P = π −1 (0, 0). π is a d-sheeted generic cover branched over {x an = y bm } and its monodromy ρ¯ is the composition ρ ◦ (fa,b )∗ . Proof. By definition of fiber product we get immediately that S is analytic, and π a dsheeted generic cover branched over {x an = y bm }. The two maps π and fa,b are coverings of C2 whose branching loci intersect just in the origin. This easily implies that S is smooth outside P . Now consider the homomorphism (fa,b )∗ : π1 (C2 \ {x an = y bm }) → π1 (C2 \ {x n = m y }). Of course (fa,b )∗ (γi ) = γ[i]

0 < i 6 an,

where [i] is the remainder class of i mod n, and if we choose the correct enumeration for the points in the fibers over the two base points, we have that ∀ 0 < i 6 an, γi and (fa,b )∗ (γi ) act in the same way and this holds for any γ ∈ π1 (C2 \ {x an = y bm }). This proves the second part of the theorem. 2 Theorem 2.2. In the above hypotheses, if moreover S is a normal surface, then S is normal too. Proof. First note that we can assume b = 1. Define A = C(S) =

C{x 0 , y, z1 , . . . , zn } , I

A = C(S) =

C{x, y, z1 , . . . , zn } ψ∗I

and denote the quotient fields by Q = Quot(A),

Q = Quot(A).

ψ ∗ injects A in A and Q in Q (ψ ∗ (x 0 ) = x a ); so, we can consider A and Q as extensions of A and Q, respectively.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

9

Recall that, by definition, S normal means that A is integrally closed in Q, i.e., for any f ∈ Q such that ∃ p ∈ A[t] monic with p(f ) = 0 ⇒ f ∈ A. We must check that the same property holds for A in Q. Let f ∈ Q, p = t n + an−1 t n−1 + · · · + a0 ∈ A[t] monic such that p(f ) = 0. ω = e2πi/a acts on A (respectively Q) via the natural map  ω f (x, y, z1 , . . . , zn ) = f (ωx, y, z1 , . . . , zn ) and we have that Fix(ω)= A (respectively Q). For every l such that 0 6 l < a define X ωi (f )ωil . fl = 06i
Remark that ω(fl ) = ω

 X 06i
 X fl ωi (f )ωil = ωi+1 (f )ωil = l . ω 06i
In particular, if gl = fl x l , gl ∈ Q. ∀ 0 6 i, l < a define pli = t n + ωi (an−1 )ωil t n−1 + · · · + ωi (a0 )ωnil ; pli (ωi (f )ωil ) = 0. Q Let ql = pli , hl1 , . . . , hlan its roots in some suitable extension of Q and let rl = Q l l 16i1 <··· 0, gl is 0 in S ∩ {x = 0}, i.e., is 0 as function in S ∩ {x 0 = 0}. Thus, gl /x 0 is a holomorphic function on S \ Sing(S), then gl /x 0 is a holomorphic function on S (see [11, Chapter 6, Proposition 4]), i.e., gl /x a ∈ A. So gl gl fl = l = a x a−l ∈ A. x x We conclude the proof noting that 1 X fl . 2 f= a 06l
Corollary 2.3. If π : S → C2 is a d-sheeted generic cover branched over {x n = y m } with S a normal surface singular in P = π −1 (0, 0), ρ is its monodromy, and fa,b : C2 → C2 is the map defined above then π : S = S ×C2 C2 → C2 is the normal d-sheeted generic cover associated to the monodromy ρ = ρ ◦ (fa,b )∗ . Remark that base change via fa,b induces a partial ordering among generic covers; call a generic cover minimal if it cannot be induced by other covers via one of these base changes.

10

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Proposition 2.4. (fa,b )∗ : π1 (C2 \ {x an = y bm }) → π1 (C2 \ {x n = y m }) and ψ∗ : π1 (S \ {P }) → π1 (S \ {P }) are surjective. Proof. (fa,b )∗ is surjective since (fa,b )∗ (γi ) = γ[i] for 0 < i 6 an, so it sends the minimal standard generators onto the minimal standard generators. Call R = π −1 (Cn,m ), R = π −1 (Can,bm ); π1 (S \ R) is the subgroup of π1 (C2 \ Cn,m ) of those γ such that ρ(γ )(1) = 1, and the same holds for π1 (S \ R) in π1 (C2 \ Can,bm ) and ρ. Since (fa,b )∗ is surjective and ρ = ρ ◦ (fa,b )∗ , ψ∗ : π1 (S \ R) → π1 (S \ R) is surjective too. Considering the following commutative diagram π1 (S \ R)

ψ∗

π1 (S \ R)

π1 (S \ {P })

ψ∗

π1 (S \ {P })

we obtain that ψ∗ : π1 (S \ {P }) → π1 (S \ {P }) is surjective. 2 Our aim is to classify all homomorphisms ρ : π1 (C2 \ Cn,m ) → Sd (up to inner automorphisms) whose image is transitive and such that τi = ρ(γi ) is a transposition (i.e., homomorphisms representing normal surfaces, see Section 0). Observe that if the second condition holds, the image of ρ is transitive if and only if ρ is surjective. The case d = 2 is trivial since ρ is unique and gives the double cover of C2 branched over Cn,m obtained projecting on {z = 0} the surface in C3 z2 = x n − y m (note that a hypersurface with isolated singularities is normal (see [5])), so suppose d > 3. Since n

γi = Γ γi Γ

−n n/(n,m)

n

for all i (actually Γ ), it follows that ρ(Γ ) must be in the center of Sd which is n trivial if d > 3, so it must be ρ(Γ ) = 1. n Now, if n and m are both odd, ρ(Γ ) is an odd permutation and thus it cannot be equal to 1, so nm must be even. A surjective homomorphism ρ 0 : Fr → Sd from a free group Fr with r generators g1 , . . . , gr such that the image of each generator is a transposition, can be represented by a connected graph with d labeled vertices and r labeled edges in the following way: take a vertex for every l = 1, . . . , d and if ρ 0 (gh ) = (i, j ) connect the vertex i to the vertex j with the edge labeled h. Note that the same graph with the numeration of the vertices suppressed represents ρ 0 up to inner automorphisms of Sd . A permutation σ ∈ Sd acts on the set of graphs with d labeled vertices and r labeled edges in the following way: if N is such a graph then if the edge labeled l in N connects the vertices h and k then the edge labeled l in σ (N) connects the vertices σ (h) and σ (k).

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

11

Since our presentations have the peculiar forms γi = Γ γi+m Γ

−1

,

µi = Mµi+n M −1

we can interpret our monodromy ρ as a map ρ 0 : Fn (respectively Fm ) → Sd such that τi = ρ 0 (Γ )τi+m ρ 0 (Γ )−1

(respectively τi = ρ 0 (M)τi+n ρ 0 (M)−1 ),

i.e., by a connected graph N with d vertices and n (respectively m) labeled edges such that if we act on N by ρ 0 (Γ ) (respectively ρ 0 (M)) for a (fixed) numeration of the vertices, then the edge labeled by j is transformed into the edge labeled by j + m with cyclical indices (mod n) (respectively j + n with cyclical indices (mod m)). By sake of simplicity, from now on we consider only the presentation of π1 (C2 \ Cn,m ) in terms of minimal standard generators and so graphs with n edges. Observe that in order to have a connected graph it must be d 6 n + 1. So, a generic cover branched over {x n = y m } is defined by a graph with d vertices and n labeled edges, and an integer m. Remark that we have proved that (fa,b )∗ acts on the graphs substituting the edge labeled j with a edges labeled j + sn for 0 6 s < a, and multiplying m by b. So we restrict ourselves to “minimal” monodromy graphs, i.e., graphs associated to minimal covers. Note that we are interested in graphs with labeled edges up to a cyclical permutation of the edges as we can see from the third presentation in Proposition 1.1. 3. Generic covers branched over {x n = y m } with n, m relatively prime The aim of this section is to classify all edge labeled graphs corresponding to generic monodromies ρ : π1 (C2 \ Cn,m ) → Sd in the case n and m are relatively prime and d > 3. Definition 3.1. Given a graph N , p a vertex of N and L an edge of N , let the valence of p be the number of edges of N having p as an end point, and let the valence of L be the number of edges of N with the same end points as L. Definition 3.2 (Polygons). Let a polygon with d vertices, valence a and increment j , with j and d relatively prime, be a graph with n = ad labeled edges of valence a, d vertices of valence 2a and such that ∀s, t the edges labeled s and t have (1) two vertices in common if and only if s − t = λd, (2) one vertex in common if and only if s − t = λd + j or s − t = λd − j , (3) no vertices in common otherwise. Definition 3.3 (Double stars). Let a double star of type (j, k), with j and k relatively prime, and valence a be a graph with n = aj k labeled edges of valence a, d = j + k vertices of which j of valence ak and k of valence aj and such that ∀s, t the edges labeled s and t have (1) two vertices in common if and only if s − t = λj k, (2) one vertex in common if and only if s − t = λj k + µj or s − t = λj k + µk, (3) no vertices in common otherwise.

12

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Fig. 1. A polygon with 5 vertices, valence 3 and increment 2.

Fig. 2. A double star of type (3, 4) and valence 1.

Observe that between each vertex of valence aj and each vertex of valence ak there are exactly a edges and that there are no edges between vertices of the same valence. Recall that, in both cases, the labeling of the edges is cyclical (mod n). The main result of this section is the following Theorem 3.4. The graphs which correspond to generic monodromies ρ : π1 (C2 \ Cn,m ) → Sd for d > 3 with (n, m) = 1, are the following: (1) Polygons with d vertices, valence n/d (or valence m/d) and increment j , with (j, d) = 1, j < d/2, j (d − j )|m (respectively j (d − j )|n). Moreover, d must divide n (respectively m). (2) Double stars of type (j, d − j ) and valence n/j (d − j ) (or valence m/j (d − j )), with (j, d) = 1, j < d/2, j (d − j )|n (respectively j (d − j )|m). Moreover, d must divide m (respectively n). The base change induced by the map f (x, y) = (y, x) in C2 takes graphs of type 1 in graphs of type 2, and vice versa. Let N be the graph representing the monodromy ρ. N has n edges and d vertices. Since n and m are relatively prime, the inner action of the subgroup generated by Γ on the γi is transitive, and so is the action of ρ(Γ ) (and its powers) on the edges of N .

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

13

Thus all edges have the same valence v and v|n. Call a petal the set of all edges between two fixed vertices. If v > 2 then choose two edges, say i and i + h, in the same petal. With a suitable power of Γ you can send the edge i in the edge i + h, but if so, since then the set of the vertices of the petal is fixed, the petal remain fixed (as a set), so the edges i + 2h (image of the edge i + h), i + 3h, . . . , are in the petal. If another edge, say i + k, is in the same petal but h6 | k, then all edges i + (h, k), i + 2(h, k), . . . , are in the petal. Thus, in general, the v edges in a petal are labeled i, i + n/v, i + 2n/v, . . . , i + (v − 1)n/v. Observe that you can retrieve the numeration of the whole graph once you know a suitable numeration for the case of edges of valence 1, so assume v = 1. (This corresponds to studying minimal covers in the sense of Section 2.) Since ρ 0 (Γ ) acts by conjugation, it transforms relations between the ρ 0 (γi )’s in relations of the same form between the ρ 0 (Γ )ρ 0 (γi )ρ 0 (Γ )−1 ’s, thus intersecting edges go to intersecting edges and non intersecting edges go to non intersecting edges. Moreover, the valence of a vertex is maintained, thus there are only two possible valences for the vertices (possibly equal) and each edge has vertices of both valences. Call an end a vertex of valence 1, and a leaf an edge with an end as vertex. Lemma 3.5. If N has a leaf then N is a double star of type (1, d − 1) and valence 1; moreover d|m. Proof. If N has a leaf then every other edge is a leaf. In this case d = n + 1, there is a vertex of valence n and there is only one possible numeration of the edges, i.e., N is a double star of type (1, n) and valence 1.

In order to calculate m we must construct Γ . Let Γr,s = (γ1 · · · γn )r γ1 · · · γs for r > 0, 0 6 s < n and act on the vertices of the edge labeled 1 first by Γ0,1 = γ1 , then by Γ0,2 = γ1 γ2 , . . . , Γ1,0 = γ1 · · · γn , Γ1,1 = γ1 · · · γn γ1 and so on until for Γr¯,¯s the two vertices coincide with the vertices of another edge, say the edge 1 + k. Then check if k = s¯ and if ρ 0 (Γr¯ ,¯s ) sends the edge i in the edge i + s¯ . If this is the case then Γ = Γr¯α,¯s and m = α(n¯r + s¯). It is easy to see that in this case r¯ = 1, s¯ = 1, so Γ = (γ1 · · · γn γ1 )α , d|m, and N gives a homomorphism ρ : π1 (C2 \ Cd−1,αd ) → Sd . 2

14

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Fig. 3.

Fig. 4.

So you get a d-sheeted covering of C2 ramified on x a(d−1) = y bd and it can be realized as the projection on {z = 0} of the normal surface in C3 zd − dy b z + (d − 1)x a = 0. Note that this surface is smooth ⇔ a = 1. We may now assume that N has no ends (no leaves) and edges of valence 1. Consider two edges of N , say i and i + j , with a (single) vertex in common and label their vertices as in Fig. 3. Suppose that the power of ρ 0 (Γ ) that takes the edge i to the edge i + j acts on the vertices sending a 7→ b 7→ c. Lemma 3.6. In the above case N is a polygon with d vertices, valence 1 and increment j ; moreover j (d − j )|m. Proof. With the same power of ρ 0 (Γ ) that takes the edge i to the edge i + j , the edge i + j is sent to the edge i + 2j and this must contain the vertex c (Fig. 4). Proceeding in this way we obtain a sequence of edges labeled i + kj such that two edges have in common (only) one vertex if and only if they are labeled i + hj and i + (h + 1)j for some h. Eventually one of these edges must have a vertex in common with the edge i and this must be the vertex a, thus closing the circle (if not there would be an edge of valence at least 2, see Fig. 5). This last edge must have the label i − j . Observe that every edge must belong to a circle like this one with the same j . Suppose now that there is another edge, say with the label i + l, containing the vertices b and d (Fig. 6). Again a suitable power of ρ 0 (Γ ) takes the edge i to the edge i + l and suppose b 7→ d.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

15

Fig. 5.

Fig. 6.

Fig. 7.

Then the power of ρ 0 (Γ ) which sends the edge i to the edge i + j + l must send the vertex a in both the vertices c and d, a contradiction. Otherwise, suppose b fixed for the power of ρ 0 (Γ ) which sends the edge i in the edge i + l; then the edge i + j is sent to the edge i + j + l which must contain the vertex b too. Moreover the edge i + j + l must contain the vertex c as one can see acting with the power of ρ 0 (Γ ) that takes the edge i to the edge i + j (and so the edge i + l to the edge i + l + j ). This contradicts our assumption that all edges have valence 1 (Fig. 7).

16

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

So the only possibility is that there is only one circle with n = d edges labeled consecutively 1, 1 + j, . . . , 1 + hj, . . . , 1 + d − j with 1 6 j < d/2 and (j, d) = 1, i.e., a polygon with d vertices, valence 1 and increment j . To calculate m, observe that one of the vertices of the edge 1 moves once every j steps, i.e., moves once applying γ1 , twice applying γ1 · · · γ1+j and so on, while the other vertex moves once every n − j steps in the opposite direction. Under this condition they will be again consecutive vertices every l steps with     l −1 l−1 + + 2 ≡ 0 (mod d), j d −j and they are the vertices of the edge labeled       l −1 l−1 +1 j =1+ + 1 (d − j ). 1+ j d −j We require this edge to be the edge labeled l + 1, so    l−1 + 1 j + kd l= j with k > 0, thus   l−1 j −j <1 kd = 1 + j and it must be k = 0. Thus, j |l and the same argument shows that (d − j )|l. The minimal l verifies     l −1 l−1 + + 2 = d, j d −j thus l = j (d − j ). Now j (d − j )|m. In fact, since the graph does not change if we cyclically permute the edges, we have that acting on the edge labeled 2 by γ2 · · · γl+1 we obtain the edge labeled 2 + l. If j 6= 1 we get the same result if we act by γ1 · · · γl (the edge labeled 1 (respectively l + 1) does not intersect the edge labeled 2 (respectively 2 + l)), thus the edge labeled i is sent to the edge labeled l + i. If j = 1 it is immediate to verify that ∀ i, acting on the edge labeled i by γ1 · · · γl we obtain the edge labeled i − 1. 2 So, as before, we get (allowing edges of valence a) a d-sheeted cover branched over x ad = y bj (d−j ) with (j, d) = 1.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

17

Fig. 8.

We may now assume that for any couple of intersecting edges, say i and i + j , the suitable power of ρ 0 (Γ ) that sends the edge i to the edge i + j leaves the common vertex b fixed. Lemma 3.7. Under this assumption, N is a double star af type (j, d − j ) and valence 1; moreover d|m. Proof. In this case, all edges labeled i + hj pass through the vertex b (or, they generate a star with increment j and vertex b). As in the case of petals, if the edge i + k contains the vertex b and j6 | k then all edges in the form i + h(j, k) are in the star, so if a star contains l edges, then they are numbered i + h(n/ l) and every edge of the graph is in such a star with the same l. Observe that all edges that have a vertex b in common generate a star with an increment that divides n and vertex b (see Fig. 8). Since we assume that the graph has no leaves, then j 6= 1 and there must be another edge containing the vertex a. This edge is labeled, say, i + k with (k, j ) = 1, and, since we assume edges of valence 1, it does not contain the vertex b. Its other vertex A cannot be in common with the edge i + hj otherwise the power of ρ 0 (Γ ) that sends i to i + hj would send the edge i + k to the edge i + hj + k not fixing the common vertex (Fig. 9). Thus, also the edges containing the vertex a generate a star and we may suppose that it has increment k, i.e., that its edges are labeled i + h0 k with h0 = 0, 1, . . . , n/k − 1. The same happens at the vertices of the edges i + hj other than b, while in the vertex A there is a star with increment j (edges labeled i + k + hj with h = 0, 1, . . . , n/j − 1). The free vertices of this star coincide with the free vertices of the star with vertex b. The same happens with all the other edges of the star with vertex a (Fig. 10). Comparing the edges we come to kj edges and k + j vertices forming j stars with increment j and k edges, and k stars with increment k and j edges. Since there is only

18

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Fig. 9.

Fig. 10.

one way to number a star when the label of a single edge is known (as for petals), there is only one possible numeration of the edges, i.e., we get a double star of type (j, k) and valence 1. We can immediately see that j + k|m. 2 Allowing petals of valence a and choosing j < k, we get a d-sheeted cover branched over x aj (d−j ) = y bd with 2 6 j 6 d/2 and (j, d) = 1. Proof of Theorem 3.4. The three lemmas give the complete classification of the graphs as in the statement; so we have only to understand the action of f : C2 → C2 defined by f (x, y) = (y, x) on these graphs. This map sends the minimal standard generators (γi ) for π1 (C2 \ Cn,m ) into the standard generators (µi ) for π1 (C2 \ Cm,n ), and vice versa.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

19

−1 Being (see the proof of the Propositions 1.1 and 1.2) µi = Γi−1 γi−1 Γi−1 , this isomorphism can be seen on the graphs in this way: act on the edge i by ρ(γi−1 · · · γ1 ) (the edge 1 remains fixed). If you do this on m = aj (d − j ) consecutive edges of a polygon with d vertices, valence a and increment j you get a double star of type (j, d − j ) and valence a and if you do this on n = ad consecutive edges of a double star of type (j, d − j ) and valence a you get a polygon with d vertices, valence a and increment j . 2

By the bijection between monodromy graphs and generic covers, we get Corollary 0.2.

4. Local fundamental groups Let π : S → C2 be the cover branched over Cn,m constructed in the previous section from the polygon with n vertices, valence 1 and increment α. In order to see if S is singular (note that away from P = π −1 ((0, 0)) S is smooth) we must check whether π1 (S \ {P }) is trivial or not (see [10]), so we must calculate π1 (S \ {P }). Let β = n − α, b = αβ/m. Recall that (α, β) = 1, b is an integer and we may assume α < β. Let Ri = γi γi+1 · · · γi+m−1 , where all indices are taken cyclical (mod n). Then the relations defining π1 (C2 \ {x n = y m }) may be written as Ri = Ri+1 for i = 1, . . . , n − 1. In order to compute π1 (S \ {P }) consider π|S\R : S \ R → C2 \ Cn,m , where R = π −1 (Cn,m ). This is an unramified cover, and we can identify π1 (S \ R) with the subgroup of π1 (C2 \ Cn,m ) given by those elements γ such that ρ(γ )(1) = 1. We apply the Reidemeister–Shreier method (see [8]) to the Shreier set of left cosets Lj = γ1 γ1+α γ1+2α · · · γ1+(j −1)α for j = 0, . . . , n − 1 (L0 = e). A set of generators for π1 (S \ R) is given by the following elements −1 −1 −1 2 γ1+(i−2)α · · · γ1+α γ1 Ai = γ1 γ1+α · · · γ1+(i−2)α γ1+(i−1)α

for i = 1, . . . , n − 1; Bj,i = Lj γ1+iα L−1 j for i 6= j − 1, j , and j = 0, . . . , n − 1. C = γ1 γ1+α · · · γ1+(n−2)α γ1+(n−1)α , −1 −1 −1 · · · γ1+α γ1 D = γ1+(n−1)α γ1+(n−2)α

20

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

while a set of defining relators is given by rewriting in terms of the above generators the following −1 −1 Lj Lj Ri Ri+1

for all choices of i and j . Observe that π1 (S \ {P }) is obtained from π1 (S \ R) by adding the relations Ai = Bi,j = CD = e for all choices of i and j (they represent loops around all the components of R). Thus π1 (S \ {P }) is generated by C and to determine its order we have to go through the rewriting process for the relations. Theorem 4.1. With the above notations if α > 1, π1 (S \ {P }) = Z/bZ. Proof. First observe that α > 1 ⇒ α + β < αβ. Consider Lj Ri and write Ri = γi · · · γi+m−1 = λ1 · · · λn Ri0 where λl = γi+l−1 . Let h, k be integers such that λh = γ1+(j −1)α , λk = γ1+j α . Since if i 6= 1 + β   Bj,i Lj if i 6= 1 + (j − 1)α, 1 + j α,    Lj γi = Aj Lj −1 if i = 1 + (j − 1)α,     if i = 1 + j α, Lj +1 while

Lj γ1+β

then if h > k Lj Ri =

  Bj,1+β Lj    = CL0     DLn−1

if j 6= 0, n − 1, if j = n − 1, if j = 0,

 H Lj +1 λk+1 · · · λn R 0

if j 6= n − 1,

H CL λ 0 0 k+1 · · · λn Ri

if j = n − 1

i

while if h < k;  H Lj −1 λh+1 · · · λn R 0 i Lj Ri = H DL λ 0 n−1 h+1 · · · λn Ri

if j 6= 0, if j = 0

with H a word in the A’s and B’s. Observe that in the first case we pass first from Lj to Lj +1 and then to Lj +2 , Lj +3 and so on (cyclic indices (mod n)) every α steps, i.e.,

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

21

Lj Ri = K1 Lj +1 λ01 · · · λ0s = K1 K2 Lj +2 λ01+α · · · λ0s = K1 K2 K3 Lj +3 λ01+2α · · · λ0s .. . where Kh is a word in the A’s, B’s and C and Kh contains C if and only if j + h ≡ 0 (mod n). If this happens we say that rewriting Lj Ri the coset index increases. Analogously, in the second case we pass first from Lj to Lj −1 and then to Lj −2 , Lj −3 and so on every β steps and the corresponding Kh is a word in the A’s, B’s and D and Kh contains D ⇔ j − h ≡ n − 1 (mod n). In this case we say that the coset index decreases. Note that for a fixed i there are only α indices j for which the coset index increases, namely those indices such that 1 + j α = i, i + 1, . . . , i + α − 1. Suppose that rewriting Lj Ri the coset index increases and write 1 + j α = i + c − 1 with 0 < c 6 α. Then deleting all the words in the A’s and B’s Lj Ri = Lj +1 γi+c · · · γi+m−1 = Lj +2 γi+c+α · · · γi+m−1 .. . = CL0 γi+c+(n−j −1)α · · · γi+m−1 = C 2 L0 γi+c+(2n−j −1)α · · · γi+m−1 = C t Ll γi+c+m−α · · · γi+m−1 = C t Ll (if c = α there are no γ ’s in the last line) for a suitable l and where       bβ − (α + β) + j bβ + j m − α − (n − j − 1)α +1= +1= . t= αn α+β α+β On the other hand, suppose that rewriting Lj Ri the coset index decreases and write 1 + j a = i + c0 − 1 with α < c0 6 n. Then again Lj Ri = Lj −1 γi+c0 −α · · · γi+m−1 = Lj −2 γi+c0 −α+β · · · γi+m−1 .. . = DL0 γi+c0 −α+jβ · · · γi+m−1 = D 2 L0 γi+c0 −α+(n+j )β · · · γi+m−1 0

0

= D t Ll 0 γi+c0 +m−α−β · · · γi+m−1 = D t Ll 0 (if c0 = n there are no γ ’s in the last line) for a suitable l 0 and where     bα − j − 1 m − β − jβ 0 +1= + 1. t = βn α+β Observe that if the coset index increases rewriting both Lj Ri and Lj Ri+1 then 1 < c 6 α and

22

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Lj Ri+1 = Lj +1 γi+c · · · γi+m = Lj +2 γi+c+α · · · γi+m .. . = CL0 γi+c+(n−j −1)α · · · γi+m = C 2 L0 γi+c+(2n−j −1)α · · · γi+m = C t Ll γi+c+m−α · · · γi+m = C t Ll , −1 −1 Lj for which the coset index increases for both Lj Ri that is rewriting a relation Lj Ri Ri+1 and Lj Ri+1 yields the trivial relation C t = C t . The same thing happens if the coset index decreases rewriting both Lj Ri and Lj Ri+1 , 0 0 i.e., if α < c0 6 n and we get D t = D t . In the case c = 1, (respectively c0 = α + 1) rewriting Lj Ri the coset index increases (respectively decreases) and rewriting Lj Ri+1 the coset index decreases (respectively 0 0 increases) so we get C t = D t , i.e., C t +t = 1. Observe that t + t 0 = b, in fact write

bβ = r(α + β) + s, bα = r 0 (α + β) + s 0 , with 0 6 s, s 0 < α + β. Adding the above equations we get b(α + β) = (r + r 0 )(α + β) + s + s 0 , i.e., (α + β)|(s + s 0 ) which implies s + s 0 = 0 or s + s 0 = α + β. In the first case we have s = s 0 = 0 (this is true if and only if (α + β)|b) and bβ + j = r(α + β) + j , bα − j − 1 = r 0 (α + β) − j − 1 which implies t = r, t 0 = r 0 − 1 + 1 ∀ j so that t + t 0 = r + r 0 = b. In the other case (s, s 0 6= 0) we have bβ + j = r(α + β) + s + j,

bα − j − 1 = r 0 (α + β) + s 0 − j − 1

and s + j > α + β ⇔ s 0 − j − 1 = α + β − s − j − 1 < 0 so we have only two possibilities t + t 0 = r + r 0 + 1 or t + t 0 = r + 1 + r 0 − 1 + 1 that is t + t 0 = b. Summing up π1 (S \ {P }) = hC | C b = 1i ≡ Z/bZ. 2 Consider now the n = α + β-sheeted generic cover π : S → C2 branched over Ca(α+β),bαβ corresponding to the polygon with α + β vertices, valence a and increment α. This surface is obtained from S via fiber product with fa,1 (see Section 2). Corollary 4.2. If a = b = 1 S is smooth. If b 6= 1 S is singular. Proof. If αβ < α + β, i.e., α = 1 the result follows observing that in this case S ⊂ C3 has equation zn+1 − (n + 1)x a z + ny b = 0, so suppose α + β < αβ.

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

23

If a = 1 then, by the above proposition, π1 (S \ {P }) is trivial if and only if b = 1. If a > 1 then π1 (S \ {P }) 6= 0, so, by Proposition 2.4, π1 (S \ {P }) 6= 0, and S is singular. 2 5. Generic covers branched on {x n + y bn = 0} In this section S is a normal surface, π : S → C2 a d-sheeted (d > 3) generic cover branched over the curve C = {x n + y bn = 0}. In particular, b and n cannot be both odd, so bn is even. Let e S be a resolution of the (isolated) singularity of S obtained from π using the standard algorithm (see, e.g., [7]). We get the following diagram: e S

π 00

π˜

T

S π

π0

C2

with π 0 a sequence of ordinary blow-up’s, π˜ proper, π 00 a resolution of S. e S is the plumbing variety of a normal crossing configuration of smooth curves, i.e., of the compact e = π˜ −1 (π 0 −1 (C)). We look for the dual graph of this configuration (see, components of C e.g., [7]). Recall that, as in the previous sections,

π1 (C2 \ C) = γ1 , . . . , γn | Γ b γi Γ −b γi−1 , where Γ = γ1 · · · γn , with γi geometric loops around C supported on the line {y = 1} such that Γ is a loop around all the points of C on this line. Consider the monodromy ρ : π1 (C2 \ C) → Sd , and let ν be the number of orbits of σ = ρ(Γ ), k1 , . . . , kν the cardinalities of these orbits. Remark that every labeled graph with d vertices and n edges represents the monodromy of a cover branched over {x n + y bn = 0} for a suitable b; more precisely for every b multiple of the order of the permutation corresponding to the ordered product of the edges. Definition 5.1 (Strings). Call a string the dual graph of a normal crossing configuration of curves which is a tree and such that no vertex is contained in more than two edges, i.e., a graph like the following:



• · · · · · · · · · · · · · · ·•



Call a string of type Ak a string with k vertices such that every vertex corresponds to a smooth rational curve with self-intersection −2. In this section we prove the following result:

24

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Theorem 5.2. In the above hypothesis and notations, S has a (minimal) resolution which is the plumbing variety of the following normal crossing configuration of smooth curves:

where the vertex E b has genus (n − d − ν + 2)/2 and self-intersection −ν. Moreover, S is smooth ⇔ the monodromy graph is a tree; this can occur only if σ is a d-cycle, so d divides b. C0

First, we need the embedded resolution graph of C ⊂ C2 , i.e., the dual graph of = π 0 −1 (C), and the geometric loops around the irreducible components of this curve.

Lemma 5.3. The embedded resolution graph of {x n + y bn = 0} in C2 is the following:

where the vertices correspond to exceptional divisors, the number over each vertex is its self-intersection and the numbers (j n) are the multiplicities of zero of the function f 0 = π 0 ∗ f , f = x n + y bn ; the arrowhead vertices are exactly n, the number of irreducible components of C. Let Ej be the irreducible curve in T corresponding to the vertex of multiplicity j n in our graph. Let pj = Ej ∩ Ej +1 , P = (1, 1) ∈ C2 . ∀ j there exists a neighborhood Uj of pj in T , and local coordinates (ξ, η) in Uj , such that C 0 ∩ Uj = (Ej ∩ Uj ) ∪ (Ej +1 ∩ Uj ) = {ξ η = 0}, and P 0 = π 0 −1 (P ) ∈ Uj with coordinates (1, 1). Choosing P 0 as base point for π1 (T \ C 0 ), the natural geometric loops in Uj \ C 0 around Ej ({(eiθ , 1)}) and and Ej +1 ({(1, eiθ )}) are respectively Γ i and Γ i+1 under the isomorphism π 0 ∗ : π1 (T \ C 0 ) → π1 (C2 \ C). Proof. We are looking for C 0 and the monodromy around one of its irreducible components. Choose as base point the point P = (1, 1) which is not contained in C for all b and n. Consider the following lines in C2 : Π 0 = {y = 1}, Π = {x = y}, 00

Π = {x = 1}. Remark that the intersection of these lines is exactly the point P .

(1)

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

25

Fig. 11.

Let γ 0 = Γ (respectively γ¯ , γ¯ 00 ) be a loop in Π 0 (respectively Π , Π 00 ) around all points of C and let γ be a loop in Π around (0, 0) as in Fig. 11. C ∩ Π is given by the equations {x = y, x n (1 + x (b−1)n) = 0}, i.e., the origin (with multiplicity n) and (b − 1)n distinct points on the unitary circle. The pencil of lines Πλ = {x = 1 − λ + λy}, for 0 6 λ 6 1 defines an homotopy in C2 \ C between γ¯ 00 ∈ Π 00 and γ¯ ∈ Π (observe that ∀ λ C cuts on Πλ n points (with multiplicity) and P ∈ Πλ ). Now consider the lines Π λ = {y = λ}, for ε 6 λ 6 1, with 0 6 ε 6 1.  C ∩ Π λ = (λb e(2r+1)iπ/n, λ), 0 6 r 6 n − 1 . Thus, considering the paths α = {(1 − t, 1 − t)}t ∈[0,1−ε] , βl = {(ε, ε + (l − 1)εt)}t ∈[0,1] and τ = {(εeit , 0)}t ∈[0,2π] , ∀ 0 < ε < 1, γ 0 is homotopic in C2 \ C to αβ0 τβ0−1 α −1 . βλ (εeit , λεeit )t ∈[0,2π]βλ−1 for 0 6 λ 6 1 defines an homotopy in C2 \ C between β0 τβ0−1 and (εeit , εeit )t ∈[0,2π] , so γ 0 = γ in π1 (C2 \ C). C 0 is obtained by blowing up recursively C2 in points of our curves (and his complete transform). So, the complementary of the complete transform does not change after every blow-up. With a slight abuse of notation, we do not change name to paths after every blowup. The first step is to blow up C2 in the origin. Recall that we get a complex manifold obtained pasting two charts biholomorphic to C2 , respectively V1 and U1 , with projections on C2 given (in coordinates) respectively by: (V1 )

(x, y) 7→ (xy, y),

(U1 )

(x, y) 7→ (x, xy).

The complete transform of C in these two charts is, in (V1 ), y n (x n + y (b−1)n ) = 0, and in (U1 ), {x n (1 + x (b−1)ny bn ) = 0}. Remark that the last one is smooth (after reduction). Then the singularities of the complete transform of our curve are in (V1 ). V1 ∼ = C2 , so we can compute the lines Π10 , Π1 , Π100 (i.e., the lines in (V1 ) given by the Eq. (1)). Note that the inverse image of P in (V1 ) has coordinates (1, 1). We claim that in (V1 ) we have the following situation (see Fig. 12). In fact, from the explicit equation of the projection, Π10 is exactly Π 0 , while Π100 is exactly Π . Rewriting the homotopies we find that the loop around the points cut by C1 , the strict transform of C, on Π1 is γ . We have to find the homotopy class of the loop around

26

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Fig. 12.

the origin in Π1 , that is of η = α(εeit , εeit )t ∈[0,2π]α −1 where α is a real positive path (i.e., α = (α1 , α2 ) with αi (t) ∈ R+ ) from (1, 1) to (ε, ε), as a loop in C2 \ (C1 ∪ {y = 0}). Since |y| 6 1 ⇒ |x| 6 1 and |x| 6 |y| then, by the homotopy (λεeit , εeit )t ∈[0,2π] , with 1 6 λ 6 2, η ∼ η0 = β(2εeit , εeit )t ∈[0,2π]β −1 with β a real positive path from (1, 1) to (2ε, ε). Now   2εei(1+λ)t , εe(1−λ)t t ∈[0,π] ∪ 2εei(2πλ+(1−λ)t ), εei((2t −2π)λ+(1−λ)t ) t ∈[π,2π] for 0 6 λ 6 1 defines a homotopy η0 ∼ β(2εeit , ε)t ∈[0,2π](2ε, εeit )t ∈[0,2π]β −1 . Moreover   2ε + λ(1 − ε) eit , ε + λ(1 − ε) t ∈[0,2π] for 0 6 λ 6 1 gives β(2εeit , ε)t ∈[0,2π]β −1 ∼ β 0 ((1 + ε)eit , 1)t ∈[0,2π]β 0 −1 with β 0 a real positive path in {y = 1} from (1, 1) to (1 + ε, 1) and  λ + (1 − λ)2ε, εeit t ∈[0,2π] for 0 6 λ 6 1 gives β(2ε, εeit )t ∈[0,2π]β −1 ∼ β 00 (1, εeit )t ∈[0,2π]β 00 −1 with β 00 a real positive path in {x = 1} from (1, 1) to (1, ε). Thus the loop around the origin in Π1 is just γ 2 . Now we conclude the proof by induction. The inductive hypothesis is that after k blow-up’s in the origin (of the V -charts), the complete transform of C has equations, y kn (x n + y (b−k)n ) = 0 in Vk , and x kn y (k−1)n(1 + x (b−k)n y (b−k+1)n) = 0 in Uk ; moreover the complete transform of C in Ukc has as dual graph a string of type Ak−1 . One extremal component of this configuration intersects (transversally in the origin of Uk ) the new exceptional divisor, and on Vk our lines are as in Fig. 13. Call El the exceptional curve of the lth blow-up, or, with abuse of notation, its strict transform in every other blow-up. Remark that we have proven the inductive hypotheses for k = 1. 0 00 and Πk+1 are the same as Πk0 and Πk and the paths Blow up the origin of Vk . Πk+1 in Πk can be computed as in the previous step. The new equations in Vk+1 and in Uk+1 are obtained simply from the old equations in Vk (and the equations of the blowc ; we are blowing up a point q ∈ E so the up). Moreover Ek is contained in Vk+1 k k

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

27

Fig. 13.

self-intersection of Ek pass from −1 to −2, and it intersects transversally Ek−1 , since S Sk / k−1 qk ∈ j =1 Ej is a normal crossing configuration of type Ak and Ek+1 intersects j =1 Ej . transversely Ek . Remark that Uk (with the given coordinates) is the chart we are looking for around pk = Ek ∩ Ek+1 . 2 The dual graph of the standard resolution of the singularity of S depends only on b, n, d and the conjugacy class of σ = ρ(Γ ) in Sd (this follows easily from the explicit construction of S from C and ρ, see, e.g., [7,12]). From now on, we call multiplicity of a curve in e S the multiplicity of f 0 ◦ π˜ ; ∀i, −1 E i = π˜ (Ei ). Lemma 5.4. E b is an irreducible compact connected curve of multiplicity bn and genus (n − d − ν + 2)/2. Moreover, π| ˜ E b has degree d. Proof. {σi } = {ρ(γi )} is a family of transpositions generating Sd , and if d > 3, σ b = 1. We have remarked before that σ b is the monodromy of a geometric loop around Eb , then for a small neighborhood V of a generic point of Eb , π˜ −1 (V ) has d connected components, and π| ˜ π˜ −1 (V ) is a cover of degree d. π˜ |E b is a branched cover of degree d (a priori non connected), and the multiplicity of f 0 ◦ π˜ on E b is exactly the multiplicity of f on Eb , i.e., bn. The branching points are the intersection points of Eb with the other branches of the configuration, i.e., every 0 branch of C , the strict transform of C by π 0 , and Eb−1 , so we must consider n + 1 points. A geometric loop λ in E b around such a point, acts on the d sheets in the same way as a small perturbation λ0 of λ (λ0 a loop in T \ C 0 ∼ = C2 \ C). 0 The geometric loops around the points of intersection of Eb with C are the γi ’s, which act transitively on the fiber, so E b is irreducible. ˜ E b in these points is 1. Since σi is a transposition, the branching index of π| On the other hand, a geometric loop around pb−1 = Eb ∩ Eb−1 , has monodromy ρ(Γ b−1 ) = σ −1 , which is in the same conjugacy class as σ in Sd , so the branching index in this point is exactly d − ν. Thus, by Hurwitz formula, χ(E b ) = dχ(Eb ) − n − (d − ν) = d + ν − n. 2

28

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31 i

Lemma 5.5. E b−1 has ν connected components, irreducible of genus 0, E b−1 , of multiplicity (b − 1)nki . π˜ −1 (pb−1 ) is a set of ν distinct points. Moreover, (E b )2 = −d. Proof. A geometric loop around Eb−1 is Γ b−1 . Thus, π| ˜ E b−1 is a branched cover of degree ν. Now, we have just two branching points; but a small perturbation of a geometric loop around pb−1 in E b−1 acts as the identity, and a small perturbation of a geometric loop around pb−2 in E b−1 acts as ρ(Γ b−2 ) = ˜ E b−1 is a cover σ −2 = (σ −1 )2 , and they both do not connect the orbits of σ −1 . Thus π| (unramified) of degree ν with ν connected components, i.e., E b−1 is the disjoint union of i ν curves, E b−1 , biholomorphic to Eb−1 (and thus of genus 0). Moreover π˜ is ramified of i

index ki − 1 over E b−1 , and this gives us the multiplicities. Let U be a neighborhood of pb−1 with local coordinates (ξ, η) such that the curve (Eb ∪ Eb−1 ) ∩ U has equation ξ η = 0. Then the fundamental group of the complement of this curve in U is Z2 , generated by Γ b and Γ b−1 . The connected analytic covers of U \ (Eb ∪ Eb−1 ), are then classified by the subgroups of Z2 . π˜ −1 (U ) has ν connected components Vi , each associated to an orbit Λi of σ , and π| ˜ Vi is a cover of degree exactly the cardinality ki of Λi associated to the lattice in Z2 generated by (0, ki ) and (1, 0). Thus, locally the cover is (ξ, η) 7→ (ξ ki , η), and (see [7]), π˜ −1 (Eb ∩ Eb−1 ) is formed by i ν points, E b−1 ∩ E b . In order to compute (E b )2 we need only to note that the intersection product E b · (f 0 ◦ π) ˜ = 0. Then P 2 dbn dn + (b − 1)nki =− = −d. 2 Eb = − bn bn Lemma 5.6. The resolution graph of S is:

where the e Skbi are strings which depend only on b and ki . Proof. ∀ i 6 b − 1, consider E i . As in the previous lemmas, the connected components of E i are in canonical bijection with the orbits of the subgroup of Sn generated by σ i , σ i−1 , σ i+1 . But these are the orbits of σ , then E i has ν components, and each component of E i intersects only the components of E i−1 and E i+1 associated to the same orbit (or an extremal vertex of a string dominating pi or pi−1 , see [7]). In particular the Skbi are strings, and by construction they depend only on b and ki . 2

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

29

Call now Tei a tubular neighborhood of e Skbi in e S, Skbi the string obtained by e Skbi recursively contracting all the possible exceptional curves of the first kind (smooth rational curves with 2 − (E )2 (i.e., how many times in this contraction we self-intersection −1), lkbi = (E b )e b b b Sk

i

Sk

i

contract a vertex near E b ), skbi the number of vertices of Skbi . Lemma 5.7. (1) Skbi are strings of type As b ; (2) lkbi = ki − 1;

ki

(3) skbi = (b/ki )(skkii + 1) − 1; (4) s1b = b − 1.

i S1b is a string Proof. First we prove (4). In this case ki = 1 and E j → Ej is 1 : 1 ∀ j , so e S1b . isomorphic to the string it dominates, i.e., a string of type Ab−1 and S1b = e b Observe that ki divides b ∀ i, since σ = 1, and that for ki fixed, we must only prove the parts (1) and (2) for b = ki ; in fact we can compute the ki -string for any b starting from the string for b = ki , by the following argument. i First we consider E λki . ei , so E iλk dominates Eλki as a A geometric loop around Eλki acts as the identity on T i

ki -sheeted cover totally ramified over two branching points, namely, the intersections of Eλki with Eλki +1 and Eλki −1 . i

i

i

i

So, E λki is rational by Hurwitz formula and the multiplicities of E λki , E λki +1 , E λki −1 are respectively nλki , nλki (ki + 1), nλki (ki − 1). ei is associated to the lattice In a neighborhood of Eλki ∩ Eλki ±1 , the cover restricted to T (0, ki ), (1, 0), so (see [7]) the fibers over these points are made up of a finite number of points, and the part of the string that dominates Eλki is only a vertex, corresponding to a rational curve. Its self-intersection (in e Skbi ) can be computed using multiplicities, and we obtain −

nλki (ki + 1) + nλki (ki − 1) = −2ki . nλki

The other part of the string, i.e., the parts that dominate the substrings between Eλki +1 and E(λ+1)ki −1 , are exactly the e Skkii . In fact, the irreducible components of π˜ −1 (C 0 ) which dominate the Eλki +j and Eλki +j ∩ S) in this substrings are constructed in the same way Eλki +j +1 (and the normal bundles in e as the components that dominate Ej , Ej ∩ Ej +1 for b = ki , except for Eλki +1 . In this case the construction is different from the one for E1 (E0 is not defined), but a generic loop around Eλki , acts as the identity, and the result follows. Thus we get b/ki strings connected by b/ki − 1 rational vertices with self-intersection Skkii contract to Skkii . −2ki . All the substrings e

30

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

Remark that, by construction, the strings e Skbi are symmetric (because σ e σ −1 are

conjugate). Then after the contractions the self-intersection of E λki becomes −2k1 +2lkkii = −2ki + 2(ki − 1) = −2, lki does not depends on b and skbi =

b ki b ski + − 1 ki ki

(in particular, if b > ki , skbi > 0). Thus we prove (1) and (2) in the case b = ki by induction on b. For b = 1 then ki = 1 and the result follows by (4). Now assume (1) and (2) true if b¯ < b. We know that ∀ b there exists a smooth cover of degree b generically branched on the curve {x b−1 + y b(b−1) = 0}, i.e., on a curve of the class under consideration (n = b − 1). In this case we can easily check that σ is an b-cycle, i.e., ν = 1. In fact, otherwise by the inductive hypothesis we would have a tree with at least two non empty branches without exceptional curves, and then the graph cannot be contracted (i.e., the surface cannot be smooth). Now, using the previous considerations, we find that the resolution graph of this surface has the form:

where all the vertices without decoration (if they exist) are non contractible. We know a priori that this graph must be contractible. In particular the last vertex must have self-intersection −1 (no other vertex can), i.e., lki = ki − 1, and it must be rational. Moreover the string must have only smooth rational curves with self-intersection −2. 2 Proof of Theorem 5.2. By Lemmas 5.7 and 5.6, we must only check that, if b = ki then k k skii = 0, i.e., Skii is empty. Recall that by Theorem 4.1 we know π1 (S \ {p}) for some special monodromy graphs. In that theorem, the branching curves were {x α+β = y l(αβ)}, for fixed α, β, with (α, β) = 1; in that case we have d = α + β, ν = 2, k1 = α, k2 = β, π1 (S \ {p}) = Z/lZ. Assume now α = n − 1, β = 1, l = n. We get one of our curves, with b = n − 1. n−1 + s1n−1 + 1 rational By the previous lemmas, the configuration is a string of k = sn−1 curves with self-intersection −2. So S has a singularity of type Ak (see [1]). The local fundamental group of such a singularity is a cyclic group of order equal to the number of the vertices plus 1. We know that this order is l = n, then we have n − 1 vertices n−1 = l − s1n−1 − 2 = 0. and by Lemma 5.7, s1n−1 = n − 2, so sn−1 Moreover, if σ has ν > 1 orbits then the graph has no −1 curves and S cannot be smooth. So, if S is smooth, σ is a d-cycle and σ b = 1 ⇒ d|b. Moreover, if S is smooth, the genus of E b must be 0, i.e., n − d + ν = 0, n = d − ν = d − 1. Conversely, if n = d − 1 the genus of E b is 0 and the monodromy graph has d vertices and n = d − 1 edges, i.e., it is a tree. Thus σ is transitive, ν = 1, and the resolution graph is a string of rational curves with self-intersection −2 except for the last one which has self-intersection −1. So S is smooth. 2

S. Manfredini, R. Pignatelli / Topology and its Applications 103 (2000) 1–31

31

Acknowledgements The second author would like to thank Professor Fulvio Lazzeri for introducing him to the subject, and for pointing out this problem. Both authors would like to thank Professor Fabrizio Catanese for the several useful and interesting conversations on the topic. References [1] V.I. Arnold, Dynamical Systems VI—Singularity Theory I, Encyclopaedia of Mathematical Sciences 6, Springer, Berlin, 1991. [2] E. Artin, Theory of braids, Ann. of Math. 48 (1947) 101–126. [3] J. Birman, Braids, Links and Mapping Class Groups, Princeton Univ. Press, Princeton, NJ, 1975. [4] E. Brieskorn, H. Knörrer, Plane Algebraic Curves, Birkhäuser, Basel, 1986. [5] G. Fischer, Complex Analytic Geometry, Lecture Notes in Math., Springer, Berlin, 1976. [6] R.C. Gunning, H. Rossi, Analytic Functions of Several Complex Variables, Prentice-Hall Series in Modern Analysis, Prentice-Hall, Englewood Cliffs, NJ, 1965. [7] H. Laufer, Normal Two-Dimensional Singularities, Ann. of Math. Stud. 71, Princeton Univ. Press, Princeton, NJ, 1971. [8] W. Magnus, A. Karrass, D. Solitar, Combinatorial Group Theory, Interscience, John Wiley and Sons, New York, 1966. [9] B. Moishezon, Stable branch curves and braid monodromy, Lecture Notes in Math., Vol. 862, Springer, Berlin, 1981, pp. 107–192. [10] D. Mumford, The topology of normal singularities and a criterion for simplicity, Inst. des Hautes Études Scientifiques, Publ. Math. 9 (1961) 5–22. [11] R. Narasimhan, Introduction to the Theory of Analytic Spaces, Lect. Notes in Math., Vol. 25, Springer, Berlin, 1966. [12] R. Pignatelli, Singolarità di superfici algebriche, Tesi di laurea, Università di Pisa, 1994. [13] E.R. Van Kampen, On the fundamental group of an algebraic curve, Amer. J. Math. 55 (1933) 255–260. [14] O. Zariski, On the problem of existence of algebraic functions of two variables possessing a given branch curve, Amer. J. Math. 51 (1929) 305–328.